Platon Chebotaev Τhe issue with getting nan values when using Leave-One-Out (LOO) cross-validation in GridSearchCV likely stems from a couple of potential problems related to the way GridSearchCV interacts with LeaveOneOut and the scoring method being used. Let's go through these issues step-by-step and provide a solution.

Potential Issues

Steps to Resolve

Here is how you can adjust your code:

Revised Code Example

from sklearn import datasets

from sklearn.model_selection import LeaveOneOut, GridSearchCV

from sklearn.linear_model import LinearRegression

from sklearn.metrics import make_scorer, mean_squared_error

import numpy as np

# Example dataset

X, y = datasets.load_diabetes(return_X_y=True)

# Model

lm = LinearRegression()

# Parameters for GridSearchCV

parameters = {'fit_intercept': [True, False], 'normalize': [True, False], 'copy_X': [True, False]}

# Custom scoring function

scorer = make_scorer(mean_squared_error, greater_is_better=False)

# Leave-One-Out cross-validation

loo = LeaveOneOut()

loo_lm = GridSearchCV(lm, parameters, cv=loo, scoring=scorer)

# Fit the model

loo_lm.fit(X, y)

# Get the best score and best parameters

print("Best Score: ", loo_lm.best_score_)

print("Best Parameters: ", loo_lm.best_params_)

Explanation